intro
Purpose:
To compare univariate and multivariate (multivertex) measures in terms of how well they discriminate subjects.
Description:
- Beta contrast patterns (\(\text{high} - \text{low}\)) were estimated for each run
- with these contrast patterns, two types of euclidean distances were estimated:
- across-run, within-subject distances
- across-run, betwen-subject distances
- the mean within-subject distance was subtracted from the mean between-subject distance
- this formed the “multivariate” intersubject discrimintion index (IDI)
- the same procedure was conducted on beta contrast means, to form the “univariate” IDI
- these IDIs were tested against zero, and were also directly contrasted between univariate and multivariate.
Inferential statistics:
- non-parametric bootstrap tests were conducted:
- subjects resampled, and IDIs recalculated to form bootstrapped distributions
- 95% intervals were calculated on bootstrapped distributions with BCa method
- frequentists p-values derived from percentile bootstrap
- all p-value corrections were performed whole-cortex, with FDR.
Parcellation:
GLMs:
- DMCC55B
- run-wise, 1trpk, shifted TENTs
- Cuedts: terms for \(\text{trial-type (congr, incon)}\times\text{sequence (switch, repeat)}\) instead of \(\text{incentive}\)
Target TRs:
- Axcpt: 7, 8, 9
- Cuedts: 9, 10
- Stern: 11, 12
- Stroop: 2, 3, 4
Contrasts:
- Axcpt: \(\text{BY} - \text{BX}\)
- Cuedts: \((\text{InConSwitch} + \text{InConRepeat} - \text{ConSwitch} - \text{ConRepeat})/2\)
- Stern: \(\text{LL5RN} - \text{LL5NN}\)
- Stroop: \((\text{PC50InCon} + \text{biasInCon} - \text{PC50Con} - \text{biasCon})/2\)
Transforms:
- “scaled”: beta coefficients were divided by their root mean sum of squares prior to IDI procedure.
- non-prewhitened patterns